A Medically Assisted Model for Precise Segmentation of Osteosarcoma Nuclei on Pathological Images

Area of Science:

• Medical image analysis
• Artificial intelligence in oncology
• Digital pathology

Background:

• Osteosarcoma diagnosis relies heavily on pathological images, but lacks expert pathologists in underdeveloped regions.
• Existing image segmentation methods struggle with staining variations and limited data.
• Accurate and efficient osteosarcoma diagnosis is critical, especially in resource-limited settings.

Purpose of the Study:

• To develop an intelligent assisted diagnosis and treatment scheme (ENMViT) for osteosarcoma pathological images.
• To address challenges of data scarcity and staining variations in osteosarcoma image analysis.
• To improve diagnostic accuracy and efficiency for osteosarcoma in underdeveloped areas.

Main Methods:

• Utilized KIN for image normalization on limited GPU resources.
• Employed traditional data augmentation techniques (cleaning, cropping, mosaic, sharpening) to address data scarcity.
• Developed a multi-path semantic segmentation network combining Transformer and CNN, incorporating edge offset loss and connected domain filtering.

Main Results:

• The ENMViT scheme demonstrated robust performance across all stages of osteosarcoma pathological image processing.
• Achieved a 9.4% higher IoU (Intersection over Union) index in segmentation results compared to existing models.
• Successfully filtered noise based on connected domain size, improving segmentation quality.

Conclusions:

• ENMViT offers a valuable solution for osteosarcoma pathological image analysis, particularly in resource-limited environments.
• The proposed method effectively handles staining variations and data limitations.
• ENMViT shows significant potential for improving diagnostic accuracy and efficiency in the medical industry.